DETECTION OF HOT SPOTS IN HETEROGENEOUS SOLAR PV PANELS USING DIFFERENT MACHINE LEARNING ALGORITHMS
No Thumbnail Available
Date
2024-07-04
Authors
Odeh, Sabri
Journal Title
Journal ISSN
Volume Title
Publisher
An-Najah National University
Abstract
Localized overheating, also known as hot spotting, can occur in specific areas of a solar panel where excessive heat is generated within photovoltaic (PV) systems. This problem may occur as a result of uneven distribution of current, inconsistencies in shading, issues with soldering, or failures in the packaging, leading to a condition of reverse bias and localized heating. The presence of hot spots not only accelerates the degradation of PV systems but also increases the risk of permanent damage to the panels. Therefore, it is crucial to promptly identify and rectify any hot spot faults to ensure the safe and reliable operation of the PV system. This study investigates the utilization of six distinct machine learning classifiers (SVM, KNN, DT, ANN, XGBoost, CatBoost) for the identification and categorization six cases (healthy, one hotspot, two hotspot series, two hotspots parallel, three hotspot series and more, and three hotspots parallel and more) of early hotspots in PV modules. The data collection was done by real measuring instruments such as: current and voltage sensors, also environmental sensors for irradiance and temperature. Before the implementation of the machine learning tool, the data of each examined PV module was applied by preprocessing data methods such as: cleaning, deleting outliers, and transforming raw data into a format suitable for training and evaluating machine learning models. These data focusing on series and parallel of hotspots faults not just on a known fault. Among these classifiers, the Artificial Neural Network (ANN) emerges as the most suitable machine learning technique for early detection of PV hot-spots, exhibiting an impressive detection accuracy of 97% when appropriate data samples are employed. The proposed model can be implemented to increase the efficiency of monitoring and fault diagnosis of PV panels.